Noise reduction method, program product, and apparatus

摘要

A probability model represented as the product of the probability distribution of a mismatch vector g (or clean speech x) with an observed value y as a factor and the probability distribution of a mismatch vector g (or clean speech x) with a confidence index β for each band as a factor, executes MMSE estimation on the probability model, and estimates a clean speech estimated value x^. As a result, each band influences the result of MMSE estimation, with a degree of contribution in accordance with the level of its confidence. Further, the higher the S/N ratio of observation speech, the more the output value becomes shifted to the observed value. As a result, the output of a front-end is optimized.

主权项

1. A noise reduction method comprising:
a step of generating a confidence index for each band on the basis of a spectrum of observation speech; a step of generating a probability model represented as a mixture multi-dimensional normal distribution having a dimension for each band, each normal distribution being represented as a product of a first normal distribution and a second normal distribution; and a step of estimating a mismatch vector estimated value by executing MMSE estimation on the probability model, and deriving a clean speech estimated value on the basis of the mismatch vector estimated value, wherein the first normal distribution is a probability distribution of a mismatch vector generated based on the observation speech, and wherein the second normal distribution has a zero mean, and a variance defined as a function that outputs a smaller value as the confidence index becomes greater.